Generiranje koherentnih i zanimljivih priča pomoću strojnog učenja

master's thesis

• Title (english): Generating Coherent and Engaging Stories using Machine Learning
• Author: Dejan Marinović
• Mentor: Marina Bagić Babac (mentor)
• Committee member: Marina Bagić Babac (predsjednik povjerenstva)
• Committee member: Tomislav Kos (član povjerenstva)
• Committee member: Nikica Hlupić (član povjerenstva)
• Granter: University of Zagreb; Faculty of Electrical Engineering and Computing; Zagreb
• Defense date and country: 2023-09-20, Croatia

Abstract

U nedavnoj renesansi ogromnih jezičnih modela i modela za generiranje teksta koja traje i dan danas, ove tehnologije koriste se za rješavanje velikog broja problema uključujući i zadatka automatskog generiranja priča. Ovo potpodručje obrade prirodnog jezika vuče korijene još iz samih početaka polja umjetne inteligencije te svoj razvitak nastavlja i dalje kako nastaju nove metode za automatsko generiranje priča. Iako su ovi pristupi opisani i predstavljeni u istraživačkim radovima dostupnima na Internetu, rekreirati i trenirati ove modele nije jednostavna zadaća pogotovo kada se koriste velike količine podataka koje zahtijevaju veliku računalnu moć i resurse koje prosječno računalo nema. Stoga je konstruiran sustav za automatsko generiranje priča u obliku grafičkog sučelja za upotrebu kojega korisnici ne trebaju ništa trenirati samostalno, nego mogu koristiti već spremne modele za generiranje. Korisnici mogu odabrati koji model žele iz skupa ponuđenih modela koji su trenirani na WikiPlots skupu podataka te također mogu ograničiti proces generiranja odabirom željene duljine teksta te unosom početnog ulaza.

Abstract (english)

In the recent reinassance era of large language models and text generation models, these technologies are employed for many tasks including automatic story generation. Dating back to early days of artificial intelligence, this subfield of natural language processing continues to be developed as new methods for automatic story generation are being invented. Although these approaches are described and presented in research papers available online, recreating these models and training them for usage is not a trivial task, especially when using large amounts of training data which requires great amounts of computing power and resources which an average computer lacks. Thus, a system for automatic story generation in the form of a graphical user interface is constructed for the use of which users do not need to train anything by themselves and can use these text generation models for inference. Users may choose between models which were trained for story synthesis on the WikiPlots dataset and can also constrain the generation process by controlling generation length and choosing a seed or prompt input.

• Keywords: generiranje priča; umjetna inteligencija; strojno učenje; obrada prirodnog jezika; duboko učenje; GPT; LSTM; WikiPlots
• Keywords (english): story generation; artifical intelligence; machine learning; natural language processing; deep learning; GPT; LSTM; WikiPlots
• Language: english
• URN:NBN: urn:nbn:hr:168:762895
• Study programme: Title: Computing; Study programme type: university; Study level: graduate; Academic / professional title: magistar/magistra inženjer/inženjerka računarstva (magistar/magistra inženjer/inženjerka računarstva)
• Type of resource: Text
• File origin: Born digital
• Access conditions: Closed access
• Created on: 2023-10-11 09:41:28